Unsupervised Damage Detection for Offshore Jacket Wind Turbine Foundations Based on an Autoencoder Neural Network.

Area of Science:

• Structural Health Monitoring
• Offshore Engineering
• Artificial Intelligence

Background:

• Offshore wind turbine foundations, particularly jacket-type, are critical for fixed wind farms in deep waters.
• Current damage diagnosis methods struggle with limited failure data, unknown environmental conditions, and diverse operational inputs.
• Existing strategies often fail when historical fault data is unavailable.

Purpose of the Study:

• To develop a robust damage diagnosis strategy for offshore wind turbine jacket foundations.
• To address the limitations of existing methods by utilizing only healthy operational data.
• To enable monitoring under varying environmental and operational conditions using vibration data.

Main Methods:

• Implementation of an autoencoder neural network model for damage diagnosis.
• Training the model exclusively on healthy operational data.
• Utilizing vibration data from accelerometer sensors for condition monitoring.

Main Results:

• The proposed strategy successfully diagnoses damage using only healthy data.
• The autoencoder model demonstrates effectiveness across different operating and environmental conditions.
• Experimental validation was performed on a scaled model in laboratory tests.

Conclusions:

• The autoencoder-based strategy offers a reliable approach for structural health monitoring of offshore wind turbine foundations.
• This method overcomes the challenge of limited or absent fault data.
• It provides a versatile solution adaptable to diverse offshore operational environments.